MakeoverMonday: AI Workslop - The Hidden Cost of AI-Generated Noise

MakeoverMonday

Python

Productivity

Workers spend 4.5 hours/week cleaning up AI mistakes - Finance teams hit hardest with 85% reporting negative consequences

Author

chokotto

Published

April 7, 2026

Overview

AI tools promised to save us time. And they do – 92% of workers say AI boosts their productivity. But there is a hidden cost: workslop, the low-effort AI-generated output that looks polished but lacks substance, dumping cognitive work onto whoever has to deal with it.

A Zapier survey of 1,100+ U.S. enterprise AI users found workers spend an average of 4.5 hours per week revising, correcting, and redoing AI outputs. That is more than half a working day – every week – cleaning up after a tool sold as time-saving.

Data Source: Zapier AI Workslop Survey (Jan 2026) / HBR: Why People Create AI Workslop (Jan 2026)
Sample: 1,100+ full-time U.S. enterprise AI users
Angle: Where workslop hits hardest, what tasks generate it, and the training paradox

Data

Show code

import pandas as pd
import numpy as np
import plotly.graph_objects as go

Show code

dept = pd.DataFrame({
    "Department": [
        "Engineering / IT / Data", "Finance & Accounting",
        "Business Development", "Operations & Supply Chain",
        "Human Resources", "Product Development",
        "Legal", "Marketing",
        "Sales & Customer Support", "Project Management",
    ],
    "Avg_Hours": [5.0, 4.6, 3.9, 3.9, 3.8, 3.8, 3.6, 3.3, 3.0, 2.8],
    "Pct_5plus": [44, 47, 29, 25, 26, 21, 20, 22, 13, 16],
    "Pct_Negative": [78, 85, 65, 59, 73, 76, 70, 78, 62, 69],
})

consequences = pd.DataFrame({
    "Consequence": [
        "Work rejected by stakeholder",
        "Privacy / security incident",
        "Customer complaint / escalation",
        "Missed deadline / delayed deliverable",
        "Compliance / legal / accuracy issue",
    ],
    "Pct": [28, 27, 25, 24, 24],
})

task_types = pd.DataFrame({
    "Task": [
        "Data analysis & visualizations",
        "Research & fact-finding",
        "Long-form reporting",
        "Email / customer communications",
        "Marketing / creative content",
    ],
    "Pct": [55, 52, 52, 46, 44],
})

training = pd.DataFrame({
    "Group": ["Untrained", "Trained"],
    "Pct_Productive": [69, 94],
    "Avg_Hours_Fixing": [2.0, 5.0],
    "Pct_5plus": [8, 44],
    "Pct_Negative": [50, 78],
})

My Makeover

Finance teams pay the steepest price for AI workslop

Show code

dept_sorted = dept.sort_values("Avg_Hours", ascending=True)
colors = [ACCENT if v == dept_sorted["Pct_Negative"].max() else MUTED
          for v in dept_sorted["Pct_Negative"]]

fig = go.Figure()
fig.add_trace(go.Bar(
    y=dept_sorted["Department"],
    x=dept_sorted["Avg_Hours"],
    orientation="h",
    marker_color=colors,
    text=[f"{v:.1f} hrs  |  {n}% negative" for v, n in
          zip(dept_sorted["Avg_Hours"], dept_sorted["Pct_Negative"])],
    textposition="outside",
    textfont=dict(size=11),
    hovertemplate="%{y}<br>%{x:.1f} hours/week<extra></extra>",
))

fig.update_layout(
    **THEME,
    title=make_title("Finance & Accounting: 85% negative consequences -- the highest of any function"),
    height=450,
    margin=dict(t=100, b=80, l=10, r=10),
    xaxis=dict(title="Average hours/week fixing AI outputs", range=[0, 7]),
    yaxis=dict(title=""),
    showlegend=False,
)

add_legend_note(fig, "Bar color: red = highest negative consequence rate among departments")
add_source(fig)
fig.show()
fig.write_image("chart-1.png", width=1200, height=600, scale=2)

Data analysis, not writing, is the biggest workslop generator

Show code

tt_sorted = task_types.sort_values("Pct", ascending=True)
tt_colors = [ACCENT if v == tt_sorted["Pct"].max() else SECONDARY
             for v in tt_sorted["Pct"]]

fig2 = go.Figure()
fig2.add_trace(go.Bar(
    y=tt_sorted["Task"],
    x=tt_sorted["Pct"],
    orientation="h",
    marker_color=tt_colors,
    text=[f"{v}%" for v in tt_sorted["Pct"]],
    textposition="outside",
    textfont=dict(size=12),
    hovertemplate="%{y}<br>%{x}% report most cleanup needed<extra></extra>",
))

fig2.update_layout(
    **THEME,
    title=make_title("55% say data analysis needs the most cleanup -- more than writing or marketing"),
    height=350,
    margin=dict(t=100, b=80, l=10, r=10),
    xaxis=dict(title="% reporting most cleanup needed", range=[0, 65]),
    yaxis=dict(title=""),
    showlegend=False,
)

add_legend_note(fig2, "Workers could select multiple task types")
add_source(fig2)
fig2.show()

74% have faced real consequences from AI outputs

Show code

cons_sorted = consequences.sort_values("Pct", ascending=True)
cons_colors = [ACCENT if v == cons_sorted["Pct"].max() else WARNING
               for v in cons_sorted["Pct"]]

fig3 = go.Figure()
fig3.add_trace(go.Bar(
    y=cons_sorted["Consequence"],
    x=cons_sorted["Pct"],
    orientation="h",
    marker_color=cons_colors,
    text=[f"{v}%" for v in cons_sorted["Pct"]],
    textposition="outside",
    textfont=dict(size=12),
    hovertemplate="%{y}<br>%{x}% of respondents<extra></extra>",
))

fig3.update_layout(
    **THEME,
    title=make_title("28% had work rejected, 27% triggered a security incident"),
    height=350,
    margin=dict(t=100, b=80, l=10, r=10),
    xaxis=dict(title="% of respondents experiencing this consequence", range=[0, 35]),
    yaxis=dict(title=""),
    showlegend=False,
)

add_legend_note(fig3, "74% experienced at least one negative consequence from AI outputs")
add_source(fig3)
fig3.show()

The training paradox: trained workers fix more but gain more

Show code

metrics = ["Say AI boosts\nproductivity", "Avg hours/week\nfixing AI", "Spend 5+ hours/week\non cleanup", "Experienced negative\nconsequence"]
untrained = [69, 2.0*10, 8, 50]
trained = [94, 5.0*10, 44, 78]

fig4 = go.Figure()
fig4.add_trace(go.Bar(
    name="Untrained",
    x=metrics,
    y=[69, 20, 8, 50],
    marker_color=MUTED,
    text=["69%", "2.0 hrs", "8%", "50%"],
    textposition="outside",
    textfont=dict(size=11),
    hovertemplate="%{x}<br>Untrained: %{text}<extra></extra>",
))

fig4.add_trace(go.Bar(
    name="Trained",
    x=metrics,
    y=[94, 50, 44, 78],
    marker_color=SECONDARY,
    text=["94%", "5.0 hrs", "44%", "78%"],
    textposition="outside",
    textfont=dict(size=11),
    hovertemplate="%{x}<br>Trained: %{text}<extra></extra>",
))

fig4.update_layout(
    **THEME,
    title=make_title("Trained workers spend 2.5x more time fixing AI -- yet 94% still say it's worth it"),
    height=400,
    margin=dict(t=100, b=80, l=10, r=10),
    barmode="group",
    yaxis=dict(title="", showticklabels=False),
    showlegend=True,
    legend=dict(orientation="h", yanchor="bottom", y=1.02, xanchor="right", x=1),
)

add_source(fig4, position="top")
fig4.show()

Key Takeaways

Workslop is a finance and engineering problem, not a writing problem. Data analysis (55%) and research (52%) create more rework than emails (46%) or marketing copy (44%).
The cost is not evenly distributed. Finance teams have the highest negative consequence rate (85%), while sales teams report the lowest (62%). The functions where errors matter most are the ones getting hit hardest.
Training is paradoxically expensive and essential. Trained workers spend 2.5x more time cleaning up AI outputs – but they also use AI in higher-stakes, higher-value contexts. The 6x productivity gap between trained and untrained workers makes the investment clear.
92% say AI is worth it despite the cleanup. The math suggests that even with 4.5 hours/week of rework, AI saves more time than it costs. But organizations without formal QA processes or context infrastructure capture significantly less of that value.

This post is part of the MakeoverMonday weekly data visualization project.

Disclaimer

This analysis is for educational and practice purposes only. Data visualizations and interpretations are based on the provided dataset and may not represent complete or current information.

--- title: "MakeoverMonday: AI Workslop - The Hidden Cost of AI-Generated Noise" description: "Workers spend 4.5 hours/week cleaning up AI mistakes - Finance teams hit hardest with 85% reporting negative consequences" date: "2026-04-07" x-posted: false author: "chokotto" categories: - MakeoverMonday - Python - AI - Productivity source-topic: "Why People Create AI Workslop" source-url: "https://data.world/makeovermonday/2026w13-ai-workslop" image: "thumbnail.svg" code-fold: true code-tools: true code-summary: "Show code" twitter-card: card-type: summary_large_image image: "thumbnail.png" title: "MakeoverMonday: AI Workslop" description: "Workers spend 4.5 hours/week cleaning up AI - Finance teams hit hardest" --- ## Overview AI tools promised to save us time. And they do -- 92% of workers say AI boosts their productivity. But there is a hidden cost: **workslop**, the low-effort AI-generated output that looks polished but lacks substance, dumping cognitive work onto whoever has to deal with it. A Zapier survey of 1,100+ U.S. enterprise AI users found workers spend an average of **4.5 hours per week** revising, correcting, and redoing AI outputs. That is more than half a working day -- every week -- cleaning up after a tool sold as time-saving. - **Data Source**: [Zapier AI Workslop Survey (Jan 2026)](https://zapier.com/blog/ai-workslop/) / [HBR: Why People Create AI Workslop (Jan 2026)](https://hbr.org/2026/01/why-people-create-ai-workslop-and-how-to-stop-it) - **Sample**: 1,100+ full-time U.S. enterprise AI users - **Angle**: Where workslop hits hardest, what tasks generate it, and the training paradox ## Data ```{python} #| label: load-packages #| message: false import pandas as pd import numpy as np import plotly.graph_objects as go ``` ```{python} #| label: load-data #| message: false dept = pd.DataFrame({ "Department": [ "Engineering / IT / Data", "Finance & Accounting", "Business Development", "Operations & Supply Chain", "Human Resources", "Product Development", "Legal", "Marketing", "Sales & Customer Support", "Project Management", ], "Avg_Hours": [5.0, 4.6, 3.9, 3.9, 3.8, 3.8, 3.6, 3.3, 3.0, 2.8], "Pct_5plus": [44, 47, 29, 25, 26, 21, 20, 22, 13, 16], "Pct_Negative": [78, 85, 65, 59, 73, 76, 70, 78, 62, 69], }) consequences = pd.DataFrame({ "Consequence": [ "Work rejected by stakeholder", "Privacy / security incident", "Customer complaint / escalation", "Missed deadline / delayed deliverable", "Compliance / legal / accuracy issue", ], "Pct": [28, 27, 25, 24, 24], }) task_types = pd.DataFrame({ "Task": [ "Data analysis & visualizations", "Research & fact-finding", "Long-form reporting", "Email / customer communications", "Marketing / creative content", ], "Pct": [55, 52, 52, 46, 44], }) training = pd.DataFrame({ "Group": ["Untrained", "Trained"], "Pct_Productive": [69, 94], "Avg_Hours_Fixing": [2.0, 5.0], "Pct_5plus": [8, 44], "Pct_Negative": [50, 78], }) ``` ```{python} #| label: shared-style #| include: false NOTE_TEXT = "Note: Zapier survey of 1,100+ U.S. enterprise AI users, Jan 2026" SOURCE_TEXT = "Source: Zapier / HBR | © 2026 chokotto" _NOTE_THRESHOLD = 50 def _build_source_note(note=NOTE_TEXT, source=SOURCE_TEXT): if len(note) <= _NOTE_THRESHOLD: return f"{note} | {source}" return f"{note} {source}" SOURCE_NOTE = _build_source_note() THEME = dict( template="plotly_white", font=dict(family="sans-serif", size=13, color="#1e293b"), paper_bgcolor="white", plot_bgcolor="#f8fafc", ) ACCENT = "#e63946" SECONDARY = "#3b82f6" POSITIVE = "#10b981" WARNING = "#f59e0b" MUTED = "#94a3b8" def make_title(text): return dict(text=text, font=dict(size=15, color="#1e293b"), x=0, xanchor="left") def add_legend_note(fig, text): margin_t = fig.layout.margin.t if fig.layout.margin.t is not None else 80 if margin_t < 100: fig.update_layout(margin=dict(t=100)) fig.add_annotation( text=text, xref="paper", yref="paper", x=0, y=1.0, xanchor="left", yanchor="bottom", showarrow=False, font=dict(size=11, color="#64748b"), align="left", ) def add_source(fig, position="bottom"): """Source annotation. position='top' places it below the title area.""" if position == "top": h = fig.layout.height or 450 mt = fig.layout.margin.t if fig.layout.margin.t is not None else 80 mb = fig.layout.margin.b if fig.layout.margin.b is not None else 80 plot_h = h - mt - mb ABOVE_PX = 15 y = 1.0 + (ABOVE_PX / plot_h) if plot_h > 0 else 1.05 fig.add_annotation( text=SOURCE_NOTE, xref="paper", yref="paper", x=0, y=y, showarrow=False, font=dict(size=10, color="#94a3b8", style="italic"), align="left", xanchor="left", yanchor="bottom", ) return TARGET_PX = 65 ANNOT_HEIGHT_PX = 35 h = fig.layout.height or 450 mt = fig.layout.margin.t if fig.layout.margin.t is not None else 80 mb = fig.layout.margin.b if fig.layout.margin.b is not None else 80 min_mb = TARGET_PX + ANNOT_HEIGHT_PX if mb < min_mb: mb = min_mb fig.update_layout(margin=dict(b=mb)) plot_h = h - mt - mb y = -(TARGET_PX / plot_h) if plot_h > 0 else -0.30 fig.add_annotation( text=SOURCE_NOTE, xref="paper", yref="paper", x=0, y=y, showarrow=False, font=dict(size=10, color="#94a3b8", style="italic"), align="left", xanchor="left", yanchor="top", ) ``` ## My Makeover ### Finance teams pay the steepest price for AI workslop ```{python} #| label: hero-dept #| fig-width: 10 dept_sorted = dept.sort_values("Avg_Hours", ascending=True) colors = [ACCENT if v == dept_sorted["Pct_Negative"].max() else MUTED for v in dept_sorted["Pct_Negative"]] fig = go.Figure() fig.add_trace(go.Bar( y=dept_sorted["Department"], x=dept_sorted["Avg_Hours"], orientation="h", marker_color=colors, text=[f"{v:.1f} hrs | {n}% negative" for v, n in zip(dept_sorted["Avg_Hours"], dept_sorted["Pct_Negative"])], textposition="outside", textfont=dict(size=11), hovertemplate="%{y} %{x:.1f} hours/week<extra></extra>", )) fig.update_layout( **THEME, title=make_title("Finance & Accounting: 85% negative consequences -- the highest of any function"), height=450, margin=dict(t=100, b=80, l=10, r=10), xaxis=dict(title="Average hours/week fixing AI outputs", range=[0, 7]), yaxis=dict(title=""), showlegend=False, ) add_legend_note(fig, "Bar color: red = highest negative consequence rate among departments") add_source(fig) fig.show() fig.write_image("chart-1.png", width=1200, height=600, scale=2) ``` ### Data analysis, not writing, is the biggest workslop generator ```{python} #| label: task-types #| fig-width: 10 tt_sorted = task_types.sort_values("Pct", ascending=True) tt_colors = [ACCENT if v == tt_sorted["Pct"].max() else SECONDARY for v in tt_sorted["Pct"]] fig2 = go.Figure() fig2.add_trace(go.Bar( y=tt_sorted["Task"], x=tt_sorted["Pct"], orientation="h", marker_color=tt_colors, text=[f"{v}%" for v in tt_sorted["Pct"]], textposition="outside", textfont=dict(size=12), hovertemplate="%{y} %{x}% report most cleanup needed<extra></extra>", )) fig2.update_layout( **THEME, title=make_title("55% say data analysis needs the most cleanup -- more than writing or marketing"), height=350, margin=dict(t=100, b=80, l=10, r=10), xaxis=dict(title="% reporting most cleanup needed", range=[0, 65]), yaxis=dict(title=""), showlegend=False, ) add_legend_note(fig2, "Workers could select multiple task types") add_source(fig2) fig2.show() ``` ### 74% have faced real consequences from AI outputs ```{python} #| label: consequences #| fig-width: 10 cons_sorted = consequences.sort_values("Pct", ascending=True) cons_colors = [ACCENT if v == cons_sorted["Pct"].max() else WARNING for v in cons_sorted["Pct"]] fig3 = go.Figure() fig3.add_trace(go.Bar( y=cons_sorted["Consequence"], x=cons_sorted["Pct"], orientation="h", marker_color=cons_colors, text=[f"{v}%" for v in cons_sorted["Pct"]], textposition="outside", textfont=dict(size=12), hovertemplate="%{y} %{x}% of respondents<extra></extra>", )) fig3.update_layout( **THEME, title=make_title("28% had work rejected, 27% triggered a security incident"), height=350, margin=dict(t=100, b=80, l=10, r=10), xaxis=dict(title="% of respondents experiencing this consequence", range=[0, 35]), yaxis=dict(title=""), showlegend=False, ) add_legend_note(fig3, "74% experienced at least one negative consequence from AI outputs") add_source(fig3) fig3.show() ``` ### The training paradox: trained workers fix more but gain more ```{python} #| label: training #| fig-width: 10 metrics = ["Say AI boosts\nproductivity", "Avg hours/week\nfixing AI", "Spend 5+ hours/week\non cleanup", "Experienced negative\nconsequence"] untrained = [69, 2.0*10, 8, 50] trained = [94, 5.0*10, 44, 78] fig4 = go.Figure() fig4.add_trace(go.Bar( name="Untrained", x=metrics, y=[69, 20, 8, 50], marker_color=MUTED, text=["69%", "2.0 hrs", "8%", "50%"], textposition="outside", textfont=dict(size=11), hovertemplate="%{x} Untrained: %{text}<extra></extra>", )) fig4.add_trace(go.Bar( name="Trained", x=metrics, y=[94, 50, 44, 78], marker_color=SECONDARY, text=["94%", "5.0 hrs", "44%", "78%"], textposition="outside", textfont=dict(size=11), hovertemplate="%{x} Trained: %{text}<extra></extra>", )) fig4.update_layout( **THEME, title=make_title("Trained workers spend 2.5x more time fixing AI -- yet 94% still say it's worth it"), height=400, margin=dict(t=100, b=80, l=10, r=10), barmode="group", yaxis=dict(title="", showticklabels=False), showlegend=True, legend=dict(orientation="h", yanchor="bottom", y=1.02, xanchor="right", x=1), ) add_source(fig4, position="top") fig4.show() ``` ## Key Takeaways 1. **Workslop is a finance and engineering problem, not a writing problem.** Data analysis (55%) and research (52%) create more rework than emails (46%) or marketing copy (44%). 2. **The cost is not evenly distributed.** Finance teams have the highest negative consequence rate (85%), while sales teams report the lowest (62%). The functions where errors matter most are the ones getting hit hardest. 3. **Training is paradoxically expensive and essential.** Trained workers spend 2.5x more time cleaning up AI outputs -- but they also use AI in higher-stakes, higher-value contexts. The 6x productivity gap between trained and untrained workers makes the investment clear. 4. **92% say AI is worth it despite the cleanup.** The math suggests that even with 4.5 hours/week of rework, AI saves more time than it costs. But organizations without formal QA processes or context infrastructure capture significantly less of that value. *** _This post is part of the [MakeoverMonday](https://www.makeovermonday.co.uk/) weekly data visualization project._ :::{.callout-caution collapse="false" appearance="minimal" icon="false"} ## Disclaimer ::: {style="font-size: 0.85em; color: #64748b; line-height: 1.6;"} This analysis is for educational and practice purposes only. Data visualizations and interpretations are based on the provided dataset and may not represent complete or current information. ::: :::